請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83302
標題: | 貝氏非循環隨機模型於傳染病監測應用 Stochastic Processes Applications to Surveillance of Infectious Disease with Bayesian Directed Acyclic Graphic (DAG) Approach |
其他標題: | Stochastic Processes Applications to Surveillance of Infectious Disease with Bayesian Directed Acyclic Graphic (DAG) Approach |
作者: | 廖翎均 Ling-Chun Liao |
指導教授: | 陳秀熙 Hsiu-Hsi Chen |
關鍵字: | 隨機過程,貝氏有向無環圖模型,新興與再現傳染病,新冠肺炎,猴痘,病毒量,臨床疾病進展, stochastic process,Bayesian Directed Acyclic Graphs(DAGs),emerging and reemerging infectious diseases,COVID-19,monkeypox,viral load,clinical disease progression, |
出版年 : | 2023 |
學位: | 博士 |
摘要: | 研究背景
新興傳染病與再現傳染病之監測,首重於在族群層次,評估疾病大規模傳播特性與流行爆發風險。另外感染病毒於個人層次所造成之動態變化,由於病毒特性與人體免疫機轉,亦對於受感染宿主在不同病毒量之下所造成的疾病傳播風險、宿主在感染後由於不同病毒量負擔所可能造成之疾病傳播與感染擴散,以及感染後產生之臨床照護需求息息相關。有效精準防治策略的發展,有賴於對前述在個人與族群層次之疾病傳播特性,以及各種防治策略,包含公衛防疫措施、族群疫苗施打策略,以及抗病毒藥物治療策略等,對於預防感染與疾病進展可達到效益之評估結果而規劃。然而傳統對於傳染病之監測評估,多著重於族群感染層次運用,如時間序列方法與傳染病傳播隔間模型,配合傳播基礎再生數指標,評估新興及再現傳染病對族群可能造成之流行風險。然而考慮前述傳染病之病毒與宿主交互影響所造成之動態變化,以及感染後所造成之疾病進展風險,傳統以族群為層次之傳染病評估架構有其限制。 研究目標 (1)在疾病傳播動態模型下拓展貝氏網絡DAG方法,建立新興與再現傳染病爆發流行之風險族群監測實證基礎。 (2)建構疾病進展多階段隨機過程貝氏網絡DAG模型運用於評估病毒以及宿主特性在個人層次對於感染後病毒量變化之影響監測。 (3)運用貝氏網絡DAG方法結合廣義線性階層模式建立大流行與區域流行評估架構,並且拓展為含括傳染病疾病進展狀態之介入效益評估方法。 材料與方法 本研究運用新冠肺炎以及猴痘全球流行為新興傳染病與再現傳染病實證資料,結合所發展之貝氏網絡模型進行實證評估。對於新冠肺炎全球流行,本研究摘取包含全球各國通報個案、住院、重症,以及死亡等公開資料。對於台灣之流行資料,本研究摘取疫苗施打以及通報個案與其發展為中症、重症,以及死亡及年齡公開資料。對於病毒量對於疾病進展之影響,本研究運用台灣某縣市之疫情調查與檢測資料。猴痘全球流行本研究運用全球各國於2022全球各國通報個案進行評估。 本研究運用貝氏網絡結合有向無環圖模型在考量傳染病傳播特性下,分別運用傳染病動態傳播模型、多階段馬可夫隨機過程,以及廣義線性迴歸模型在前述實證資料支持下,以貝氏馬可夫蒙地卡羅演算法產生前述模型中之參數事後分佈,並據以對新興傳染病以及再現傳染病之疾病傳播特性、不同防治策略,包含疫苗施打、公衛防疫措施,以及抗病毒藥物可達到之保護效果,以及感染後病毒量動態階段變化進行評估。 結果 運用貝氏網絡傳染病動態模型,結合鑽石公主郵輪侷限空間情境之新冠肺炎評估結果,顯示新冠肺炎之基礎再生數(R0)為5.70 (95% CI: 4.23-7.79)。施行鑽石公主郵輪之防疫措施對於疾病傳播可達到37% (95% CI: 33-40%)之效益。在此侷限空間情境下,若可提前5日施行防疫措施則可達到53% (95% CI: 44-62%)之疾病傳播防治效益,使新冠肺炎個案數由761人減少為403人。運用此貝氏網絡動態模型於再現傳染病猴痘2022年全球流行顯示,全球之疾病傳播再生數由初期之1.001 (95% CI: 0.986-1.150)增加至1.459 (95% CI: 1.370-1.507),隨著介入措施包含公衛防疫措施以及天花疫苗施打達到31% (95% CI: 27.0-33.6%)之介入效益,此再現傳染病之再生數下降至1.027 (95% CI: 1.026-1.026),傳播逐漸受到控制。本研究亦對各國之猴痘傳播流行再生數變化以及對應之疾病傳播防治策略進行評估。 運用貝氏網絡多階段隨機過程,本研究評估Alpha與Omicron變異株之病毒量,對於新冠肺炎感染個案由無症狀進展為臨床狀態、持續無症狀個案之比例,以及進展發生臨床症狀之時間中位數。評估結果顯示於Alpha時期多數感染者由症狀前期個案期進展成為症狀之路徑,僅有少數依循持續無症狀路徑(發生率24.9,95% CI: 15.6-35.1)。而Omicron時期依循無症狀路徑者則增加達271.4 (95% CI: 240.4-303.7)。Alpha與Omicron變異株感染發生臨床症狀之中位數時間分別為4.07天(95% CI: 3.33-4.84)以及1.22天(95% CI: 1.12-1.33)。在考慮宿主特性與接觸模式後,病毒量對於受到Alpha變異株感染個案是否產生臨床症狀以及其病程進展皆具有顯著影響,然而對於Omicron感染個案病毒量高低所造成的影響則相對減低。病毒量對於Alpha與Omicron感染後之疾病進展模式亦有所差異。病毒量於Alpha感染個案之進展呈現劑量效應,但對於Omicron感染個案病毒量影響較小,多數個案在1-2天之內由症狀前期個案期進展之發生臨床症狀。疫苗施打與否對於疾病進展亦呈現不同的模式。完整施打疫苗者受病毒量影響較小,但未施打疫苗者其感染病毒量對於其臨床動態進展則會有所影響。 運用貝氏網絡階層廣義線性迴歸DAG模型,分析新冠肺炎實證資料結果顯示,新冠肺炎之流行無法達到地方流行之平衡態,而將以四週遞延之模式受到變異株病毒傳播特性為主要影響下爆發流行。就不同防疫措施之評估結果,疫苗施打對於症狀感染可達到55% (95% CI: 38-67%)之保護效果,其中mRNA疫苗施打可提供較佳之保護效果(63%,95% CI: 40-77%),此保護效果在年輕族群(未滿50歲)優於年長族群(大於50歲)。疫苗對於死亡、重症,以及中症之保護方面,完成追加劑施打分別可提供73.9% (95% CI: 72.7-75.1%)、73.9% (95% CI: 73.0-74.8%),以及72.5 (95% CI: 71.8-73.2%)之效益,完成基礎劑之保護效益則分別為52.7% (95% CI: 49.6-55.6%)、54.5 (95% CI: 50.0-54.5%),以及51.9% (95% CI: 50.3-53.6%)。公衛防疫措施則可提供約12%-20%之保護效益。口服抗病毒藥物約可降低中症風險約12%。 結論 本研究運用隨機過程與DAG模型建立創新方法,結合多階段馬可夫模式運用以全球以及臺灣地區實證資料,發展新興與再現傳染病由族群至個人之監測架構與介入效益評估貝氏DAG階層模型。所提出多層次之傳染病監測方法有助於傳染病傳播風險完整評估以及發展個人化防治策略。 Background The evaluation for the risk of outbreak and transmission at large scale has been the first and foremost goal in the surveillance of emerging and reemerging infectious diseases at population level. In addition to this purpose, the risk of disease evolution in terms of viral shedding and clinical severity resulting from the characteristics of pathogen, viral load, and host immunity has also been an important aspect of surveillance at individual level. The development of effective and individual-tailored containment strategies including non-pharmaceutical interventions (NPIs), mass vaccination, and antiviral therapies is highly dependent on the empirical evidence taking into account these factors with multilevel characteristics. Conventional approaches for surveillance of infection disease including the use of time-series models and basic reproductive number (R0) derived from compartment models, both focusing on the risk at population level. However, the characteristics of multilevel and multiple outcomes regarding the surveillance of infectious disease render the use of these conventional approaches intractable. Aim The aims of this study are (1)to develop a Bayesian network (BN) analysis with DAG model supported by the compartment model to characterize the dynamics of disease evolution and to assess the risk of disease outbreak at population level; (2)to apply the BN DAG model with stochastic process underpinning for the surveillance of disease evolution associated with viral load and viral characteristics at individual level; (3)to develop a BN hierarchical DAG in conjunction with generalized linear regression model to assess the risk of pandemic and endemic and the effectiveness of containment strategies at different level. Materials and Methods A series of BN DAG models were developed with the support from the information contained in the empirical data of emerging (COVID-19) and reemerging (monkeypox) infectious diseases. Regarding the data on COVID-19 pandemic, global open data with the information on country and region, reported cases, hospitalized cases, cases admitted for intensive care were collected from open data. Taiwan outbreak data on the clinical severities including moderate, severe, and death, age, and vaccination history were abstracted from surveillance report. The empirical data on contact tracing and viral load measured by Ct level collected from a county in Taiwan were used for assessing disease evolution at individual level. The open data on global monkeypox outbreak surveillance in 2022 were used for assessing the risk of outbreak of re-emerging infectious disease. A series of BN DAG model in conjunction with dynamic compartment model, multistate Markov process, and generalized linear regression model taking into account disease characteristics were developed. By using the Bayesian Markov Chain Monte Carlo (MCMC) algorithm, information from the empirical data mentioned above were used for the derivation of posterior distributions of the parameters of the proposed BN DAG models. Results For the risk of COVID-19 outbreak in the confined space of cruise ship, the R¬0 derived from BN DAG susceptible, exposed, infected, and recovered (SEIR) were estimated as 5.70 (95% CI: 4.23-7.79). The effectiveness of containment measures implemented on board was estimated as 37% (95% CI: 33-40%). On the basis of the posterior distributions of the BN DAG SEIR model, the implementation of containment measured 5-days earlier can enhance the effectiveness to 53% (95% CI: 44-62%) and reduce the COVID-19 cases from 761 to 403 in such a confined space of cruise ship. Regarding the re-emerging disease of monkeypox, the R0 increased from 1.001 (95% CI: 0.986-1.150) in the early stage to 1.459 (95% CI: 1.370-1.507) in the latter stage, suggesting the risk of transmission at large scale and outbreak. The effectiveness of containment strategies including NPIs and vaccination was estimated as 31% (95% CI: 27.0-33.6%), which bring the R0 down to 1.027 (95% CI: 1.026-1.026), indicating the containment of outbreak. By using the BN DAG with stochastic process, a four-state disease progression model was constructed for the surveillance of COVID-19 evolution at individual level regarding Alpha and Omicron variants of concern (VOC) infection. The results demonstrated the difference between subjects infected by these two VOCs. While the small fraction of subjects infected by Alpha VOC turned into asymptomatic (incidence: 24.9, 95% CI: 15.6-35.1), a high incidence for asymptomatic infection (271.4, 95% CI: 240.4-303.7) was estimated for Omicron infection. The median time from pre-symptomatic to symptom phase for Alpha and Omicron VOC infection was estimated as 4.07 (95% CI: 3.33-4.84) and 1.22(95% CI: 1.12-1.33) days, a significant short period for Omicron infection. After adjusting for host characteristics and contact pattern, there was different roles of viral load for clinical evolution for Alpha and Omicron infection. A signification impact of viral load on the clinical progression for subject infection by Alpha VOC with a dose-response pattern was observed. For those infected by Omicron VOC, although viral load remains a significant effect on clinical evolution, especially for the unvaccinated population, a lower extent was observed compared with those infected by Alpha VOC The results derived by applying GLM with BN DAG show the outbreak of COVID-19 will reach equilibrium in the long run. Recurrent outbreaks affected mainly by the characteristics of dominant VOC are expected with the optimal lag function spanned over four weeks. By using the BN DAG with GLM, the effectiveness of vaccination against symptomatic infection of Omicron was estimated as 55% (95% CI: 38-67%), with a higher protective effectiveness conveyed by mRNA -based vaccine (63%,95% CI: 40-77%). Regarding the effectiveness of vaccination against death, severe, and moderate disease, the estimated results were 73.9% (95% CI: 72.7-75.1%)、73.9% (95% CI: 73.0-74.8%), and 72.5 (95% CI: 71.8-73.2%) for booster vaccination. The corresponding figures for primary series were estimated as 52.7% (95% CI: 49.6-55.6%)、54.5 (95% CI: 50.0-54.5%), and 51.9% (95% CI: 50.3-53.6%). The protective effectiveness resulting from NPIs and antiviral therapy were estimated as 12-20% and 12%, respectively. Conclusion By using Bayesian DAG approach with stochastic process underpinning, a series of novel applications to surveillance of infectious disease were developed. The proposed Bayesian DAG framework considering the effect at population level and individual level can facilitate the surveillance of emergence and re-emergence infectious disease in terms of outbreak risk assessment and the development of individual-tailored containment strategies. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83302 |
DOI: | 10.6342/NTU202300419 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 流行病學與預防醫學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-0137230211359002.pdf | 5.52 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。